What are some open research questions in AI alignment?

From Stampy's Wiki